Reports earlier this week that Roku had apparently canceled its planned voice assistants may have called the code prematurely.

The Roku voice assistant was originally announced way back in January 2018, but in the time since then, Roku hasn’t said very much about it.

As it turns out, Roku hasn’t canned its voice assistant as originally reported, but when it does finally arrive, it might not be exactly what you’re expecting.

The news that Roku had cancelled its voice assistant came in a profile about the company from The Wall Street Journal.

Since then, Roku has reached out to confirm that its efforts in voice were, in fact, still ongoing.

“We are continuing to build more voice capabilities into our platform,” Roku vice president of communications Tricia Mifsud told SlashGear, noting that those voice capabilities are called “Roku Voice.”

Apple and Samsung may have taken over the high-end market with the latest release of their flagship models (iPhone XS and Samsung Galaxy S10), but the Moto G series is the midrange market’s gamechanger.

Catch the Motorolla Moto G6 at its most affordable price yet with Amazon’s 42% price cut.

Whether you’re buying a phone for your kiddo, or just finding cost-saving alternatives, Motorola’s Moto G6 is a competent budget-friendly smartphone with acceptable performance, decent power, and capable cameras.

It may not be lauded for its durability but a Moto G6 case may just be the extra layer of protection it needs.

Compared to its previous generations and other phones within its class, the G6 proves to be visually appealing with relatively small bezels, a fingerprint sensor right below its screen, and a beautiful curved glass back.

Its 5.7-inch IPS LCD screen flaunts a 18:9 aspect ratio along with a 2,160 x 1,080 pixel resolution that makes for vibrant picture quality with vivid colors and pleasant tones.

The report found schools providing interns to work night shifts at a Hengyang Foxconn plant, producing Amazon devices.

According to the report, kids who declined compulsory overtime were told by their teachers that it would affect their graduation from school.

Foxconn - which also assembles components for Apple devices, among others - peak period begins in July.

The report claimed it has hired 1,581 pupils from local technical colleges and is looking for more.

The pupils are said to be paid $248 a month to work 10 hours a day, six days a week.

Teachers, who get a $425 bonus from Foxconn, work with factory managers to pressure the pupils into working overtime and night shifts.

Alexa is getting some much-welcome flexibility to help users who aren't digging the smart assistants automatic talking speed, after an update that gives Alexa three different speeds when responding to voice commands.

Users can now ask Alexa to "speak faster" or "speak slower", while asking the voice assistant to "speak at default speed" will bring it back to the default setting.

It looks like the feature has only gone live in the US so far – as we couldn't get a UK-based Alexa smart speaker to recreate it.

Knowing Amazon's usual update schedule, though, we expect to the feature to roll out to the UK and beyond in the coming months.

Opening up the smart home

While smart home gadgets these days can be wonderfully convenient, allowing you to switch devices on and off, change room lighting, and throw voice messages all over the house – they can also be a right hassle.

Amazon’s lead data regulator in Europe, Luxembourg’s National Commission for Data Protection, has raised privacy concerns about its use of manual human reviews of Alexa AI voice assistant recordings.

A spokesman for the regulator confirmed in an email to TechCrunch it is discussing the matter with Amazon, adding: “At this stage, we cannot comment further about this case as we are bound by the obligation of professional secrecy.” The development was reported earlier by Reuters.

Amazon’s Alexa voice AI, which is embedded in a wide array of hardware — from the company’s own brand Echo smart speaker line to an assortment of third party devices (such as this talkative refrigerator or this oddball table lamp) — listens pervasively for a trigger word which activates a recording function, enabling it to stream audio data to the cloud for processing and storage.

However trigger-word activated voice AIs have been shown to be prone to accidental activation.

So there’s always a risk of these devices recording any audio in their vicinity, not just intentional voice queries…

In a nutshell, the AIs’ inability to distinguish between intentional interactions and stuff they overhear means they are natively prone to eavesdropping — hence the major privacy concerns.

Finally, you connect the plug to your favorite home assistant, such as Amazon Alexa or Google Assistant, or the smart plug’s app.

Here are just a few to get your imagination turned on to all of the possibilities:

Turn on your coffee pot while you’re still in bed.

Control your window unit air conditioner or space heater without getting out of your chair.

Turn off the oven, curling iron, or hair straightener you left on from your desk at work.

Turn on your holiday lights using a voice command.

Amazon Alexa has tens of thousands of voice apps — skills, in Amazon’s vernacular — contributed by third-party developers.

Usually, building said skills requires supplying examples of customer requests (e.g., “Order my usual”) together with the actions to which those requests should map, which are used to train the AI system that processes real requests in production.

Needless to say, that’s labor-intensive, which is why scientists at Amazon are exploring techniques for pooling sample requests for similar requests from different skills at training time.

In a paper presented last week at the Association for Computational Linguistics conference in Florence, the coauthors write that the additional data improved performance by “plugging holes” in lists of example requests.

Evaluated on two different public corpora and an internal corpus, they report that training an AI system simultaneously on multiple skills yielded better results than training it separately for each skill.

As the researchers note, multitask training runs the risk of causing a model to lose focus on task-specific structures.